Team uses AI and satellite imagery to publish first-ever global estimates of greenhouse gas emissions from road transport

The Environmental Protection Agency estimates that the transportation sector accounts for about 27 percent of all annual greenhouse gas emissions in the United States, and emissions from road transportation, driven by carbon-generating internal combustion vehicles, in represent the vast majority.

For years, researchers have tried to measure these emissions more closely, but existing inventories are often outdated, incomplete and limited.

Now, scientists at the Johns Hopkins Applied Physics Laboratory (APL) in Laurel, Maryland, have leveraged artificial intelligence and machine learning (ML) to produce accurate estimates of road transportation emissions from all 500 cities. most emitters in the world.

“In particular, we were able to estimate the average annual daily traffic on individual road segments in urban areas and combined this with localized estimates of vehicle emissions factors to produce an estimate of total emissions,” explained Marisa Hughes, Deputy Director of the Environmental Resilience Program. in APL’s exploratory research and development mission area.

“The ability to calculate emissions by road segment provides an unprecedented level of detail and global coverage,” said Hughes, who helps manage the Lab’s climate change efforts. “The combination of ML-predicted road activity and region-specific emissions factor data creates automated, accurate, global, timely and actionable estimates of road transportation greenhouse gas emissions.”

Actionable is the key word, Hughes said. “You can’t change what you can’t measure, and that’s a big challenge when it comes to tackling climate change. We need to know where greenhouse gas emissions are coming from, but that means accounting for a lot of tiny contributions across time and geography.”

Make the invisible visible

There are approximately 1.4 billion motor vehicles in the world. An APL team of remote sensing and computer vision experts thought they could measure the greenhouse gases of these vehicles by extracting emissions signatures from visual satellite data, essentially making the invisible visible. They set about looking at existing emissions data, correlating it with those images, sorting out whatever they could with neural networks.

Then the team realized there was another way to complete the challenge, Hughes said. APL had experience mapping livelihoods for the intelligence community on a project called Functional Map of the World, where experts mapped land use and structures, and used satellite data at different times of day to understand and differentiate similar-looking structures, such as office buildings versus apartment complexes.

“We thought, what if we took that same ‘lifestyle’ analysis and instead of focusing on buildings, we looked at the networks of roads that connected them,” recalls Derek Rollend, senior ML researcher at the ‘APL. “We built a deep neural network that would take inputs available around the world: satellite imagery, road network data, and population.”

Because some regions had accurate emissions and vehicle count data, “we were able to train our machine learning models with this field data on vehicle activity, and then apply the model predictions first. at the country level, and more recently for the world’s top 500 cities,” he continued.

This work makes a major contribution to a broader effort to monitor greenhouse gas emissions globally, led by Climate TRACE [Tracking Real-time Atmospheric Carbon Emissions]an international coalition created to accelerate and facilitate meaningful climate action by independently tracking greenhouse gas emissions with unprecedented detail and speed.

This work represents how APL brings its unique expertise to make lasting contributions to the global challenges of climate change.

“It’s a great example of how our unique skills and capabilities in machine learning, satellite imagery and systems modeling can be applied to understand and one day solve complex problems,” said Bobby. Armiger, who, with Hughes, co-leads Labwide’s focus on climate change research.

On November 9, Climate TRACE released the most detailed inventory of greenhouse gas emissions ever compiled, providing asset-level emissions data for 81,087 individual sources around the world – and transportation data from ‘APL were part of this publication. In addition to publishing facility-level information, Climate TRACE has updated its independent emissions inventory for each country to include data from 2021, providing a comprehensive view of annual greenhouse gas emissions since 2015. year of the Paris Agreement, an international climate treaty. change.

Bringing radical transparency to global emissions

Members of the Climate TRACE coalition have rallied around a goal they call radical transparency.

“It’s about knowing where all the greenhouse gases are, where the emissions are coming from, and at the same time being very open about how we calculate those emissions,” Hughes said. “It’s a really big challenge. But if you have this radical transparency, you can dive in and start doing analysis and comparisons to figure out what works and what doesn’t in terms of mitigating emissions.”

One of the coalition’s goals for the coming year is to figure out how to bring together datasets and emissions inventories from different and overlapping sources to create a new best estimate of what is really happening.

“That dream of radical transparency and knowing where all the broadcasts are coming from in real time with every new satellite image is still far ahead of us,” Hughes said. “But now it feels within reach.”